Circadian Phase Estimation From Ambulatory Wearables With Particle Filtering: Accuracy Depends on Initialization,...

Area of Science:

• Chronobiology
• Wearable Technology
• Mathematical Modeling

Background:

• Mathematical models of human circadian rhythms excel in lab settings but falter in real-world applications.
• Shift workers with irregular schedules and light exposure exhibit significant discrepancies in model performance.
• The precise reasons for the reduced performance of circadian models in ambulatory settings are not fully understood.

Purpose of the Study:

• To evaluate how initialization strategy, recording duration, and light exposure characteristics affect circadian model performance.
• To assess model accuracy in predicting circadian phase from light data using wearable technology.
• To compare model performance between individuals with regular schedules and shift workers.

Main Methods:

• Utilized wearable data from individuals with regular and irregular (shift work) schedules.
• Implemented a probabilistic initialization framework to address unknown starting circadian phase.
• Assessed model performance by predicting circadian phase from light exposure data and comparing it to dim light melatonin onset measurements.

Main Results:

• For individuals on regular schedules, model accuracy improved with longer recording durations.
• Shift workers did not show accuracy gains with increased data duration, indicating a different response pattern.
• Brighter, regular light exposure correlated with better model estimates, while fragmented light exposure increased uncertainty.

Conclusions:

• Current circadian models require adaptation, particularly in light sensitivity parameters, to perform effectively in free-living, irregular conditions.
• The findings highlight the need for robust, scalable circadian tracking solutions for real-world populations.
• Irregular light exposure patterns are a key factor limiting the real-world applicability of existing circadian models.